To process extended context prompts effectively, types demand robust recall capabilities. The 'Needle In A Haystack' (NIAH) analysis steps a model's capacity to precisely remember information from the wide corpus of information. We Increased the robustness of this benchmark by using amongst 30 random needle/query pairs for each prompt and screening